Cursor 0.45: background agents autonomously fix bugs and open PRs from GitHub issues

Think of it like hiring a tireless junior developer who works overnight. You come back in the morning to find the bug fixed, tests written, and a pull request already open for your review. That is exactly what Cursor 0.45 background agents do.

Until now, even the most advanced AI coding tools required a developer to stay at the keyboard, guiding every step of the process. With this release, Cursor spins up dedicated cloud virtual machines, each running an autonomous agent assigned to a specific task. The agent reads a bug report from GitHub, explores the codebase, applies the fix, writes automated tests to verify the change, and opens a pull request — all without anyone watching.

The developer gets a notification when the work is ready for review. There is no waiting, no babysitting. Multiple agents can work in parallel on different issues at the same time.

Reaching 100,000 paying subscribers shows the market is rewarding this direction. Cursor now competes head-to-head with GitHub Copilot Workspace (backed by Microsoft and OpenAI) and Devin, the AI software engineer from Cognition that first demonstrated fully delegated coding tasks. The key difference is that Cursor arrives with an already massive developer user base and deep IDE integration.

The practical shift for developers is significant: human time moves away from writing boilerplate code and toward reviewing and validating what the agent produces. The bottleneck in software development starts to look less like writing and more like judgment.